IBIS Macromodel Task Group

Meeting date: 24 August 2021

Members (asterisk for those attending):
Achronix Semiconductor:       Hansel Dsilva
Amazon:                       John Yan
ANSYS:                      * Curtis Clark
                            * Wei-hsing Huang
Cadence Design Systems:       Ambrish Varma
                              Ken Willis
                              Jared James
Google:                       Zhiping Yang
Intel:                        Michael Mirmak
                              Kinger Cai
                              Alaeddin Aydiner
Keysight Technologies:      * Fangyi Rao
                            * Radek Biernacki
                              Ming Yan
                              Todd Bermensolo
                            * Rui Yang
Luminous Computing            David Banas
Marvell                       Steve Parker
Mathworks (SiSoft):         * Walter Katz
                              Mike LaBonte
Micron Technology:          * Randy Wolff
                              Justin Butterfield
Missouri S&T                  Chulsoon Hwang
Siemens EDA (Mentor):       * Arpad Muranyi
Teraspeed Labs:             * Bob Ross
Zuken USA:                    Lance Wang

The meeting was led by Arpad Muranyi.  Curtis Clark took the minutes.

--------------------------------------------------------------------------------
Opens:
  
- None.

-------------
Review of ARs:

- Arpad to report back to the Editorial task group on ATM's recommendation to
  use "terminal" to describe the first Tx and last Rx.
  - Done.
  
- Randy to check with the Editorial task group to see if Figure 41, the repeater
  link physical layout, should have "terminal" language added in the upcoming
  IBIS 7.1.
  - Done.  Randy reported that he had raised the issue in Editorial, but they
    had not yet decided what they will do.
  
- Fangyi to send BIRD211.3 draft 6 to ATM for review.
  - Done.
--------------------------
Call for patent disclosure:

- None.

-------------------------
Review of Meeting Minutes:

Arpad asked for any comments or corrections to the minutes of the August 17th
meeting.  Randy moved to approve the minutes.  Walter seconded the motion.
There were no objections.

-------------
New Discussion:

BIRD211.3 draft6:
Fangyi shared the draft and reviewed the changes.  As discussed in last week's
meeting:
1. Fixed the alignment of the figure representing the 4 flow cases.
2. Added "Terminal Tx" and "Terminal Rx" in the repeater link physical layout
   figure.
3. Replaced one instance of "primary Tx" with "terminal Tx".
4. Added the "impulse_matrix is presented to..." phrases and corrected grammar
   in the Redriver flows.
   
In addition, Fangyi noted that he had changed the Redriver flows so that the
AMI_Init steps for Tx1 and Rx1 stay the same as in IBIS 7.0.  These steps are
independent of Tx1's value of Tx_Impulse_Input, as is the case in the normal
flow.  Fangyi had also added the new change information for 211.3 to the
BACKGROUND INFORMATION/HISTORY section.

The group reviewed any remaining comments from previous versions.  Arpad said
he thought all of his comments had been addressed.

Curtis asked to review the second item in the Redriver flow known issues listed
in the Definition of the Issue.  It stated that the cumulative upstream impulse
response is not provided to the Redriver Tx or Redriver Rx AMI_Init functions,
but Curtis said that Rx1 does get its cumulative upstream impulse response
(the impulse response of CH-1).  Fangyi agreed, and Arpad and Radek noted that
it is only the Rx in the first Redriver that warrants an exception.  The Rx in
subsequent Redrivers would be deprived of complete upstream information (because
previous channel IRs were not necessarily optimized properly in the old flow).
Fangyi added the phrase "Except for the first Redriver Rx's AMI_Init function"
to the beginning of the sentence.

Arpad asked if anyone had additional comments.  There were none.  Bob moved that
Fangyi submit the latest draft to the Open Forum as BIRD211.3.  Radek seconded.
There were no objections.  Fangyi to send the latest version to ATM, and Randy
to post it to the Open Forum website.

PAMn BIRD213.1:
Walter said there were no new updates, but he suggested we review the existing
draft again.

AMI Reserved Parameter: Modulation_Levels
Walter said it's used to tell the tool what modulations are supported and for
the EDA tool to tell the model which one to use.  Walter asked if we want to
require that any model that supports PAMn must support NRZ (n=2).  He said he
had included this requirement because many training algorithms start off by
doing NRZ.  Arpad said we should not require all models to support NRZ.  He said
you could have a PAM4 model that might not want/need to support NRZ.  Radek
agreed with Arpad that we should not require NRZ support from every PAMn model.
Walter removed that requirement.

Walter asked whether we should allow mixing of legacy Modulation and PAM4_xxx
parameters with the new PAMn parameters being defined in this BIRD.  Arpad said
he thought we should disallow mixing the two.  He said if we avoid mixing the
two, it could make it easier to deprecate or modify parameters later without
worrying about coupling between the two sets of parameters.

Walter noted that the voltage levels of the stimulus waveform corresponding to
the different symbol values are defined formulaically in a manner similar to
that described with the earlier Modulation parameter.

Arpad asked if we had resolved the discussions about how we handle the multiple
types of certain n level modulations.  Walter said we decided that the
modulation is going to be up to the EDA tool  Fangyi agreed and said we decided
that we don't discuss 8b10b, 64b66b, etc., in IBIS, for example.  So, we will
leave this up to the EDA tool.

AMI Reserved Parameter: PAM_Thresholds
Walter said that this is an Out parameter, and it is a table so it can handle
different values of n.  He noted that he largely tried to follow the
PAM4_xxxThreshold language.  There is a similar formulaic description of how
the PAM_Thresholds and Rx_Receiver_Sensitivity are combined to define the
valid symbol voltage ranges.

Walter said it is an Out parameter so the model can tell the EDA tool what the
thresholds are.  Walter noted that he had made the parameter optional, based
on previous feedback in ATM, and the tool is required to figure them out if the
model doesn't return them.  He said, however, that model makers he spoke with
all said you can't write a PAMn model without knowing what the thresholds are
internal to the model.  The model would need them in order to figure out its
equalization settings, for example.  So, model makers seemed to think it would
be silly if the model wouldn't output thresholds.  Walter said we should decide
if we want to make this a required parameter.

Fangyi recalled that Walter had made it a required parameter in the first draft,
but someone had claimed that PAM4_[Upper|Center|Lower]Threshold parameters are
not required (Note: In IBIS 7.0, their Required: field entry says "No", but in
Usage Rules: it states that Upper and Lower are required for Rx models if
Modulation is PAM4 (Center is still optional)).  Fangyi said he agreed with
Walter's original position that PAM_Thresholds should be required.  There were
no objections, and Walter changed the language to:
   Required:  Yes, if Modulation_Levels is specified...
   
AMI Reserved Parameter: PAM_Offsets
Walter said this was not a required parameter in the current draft.  Fangyi said
he thought it should be required.  Walter said if the parameter is not specified
then there are no offsets.

Walter said there was one subtle concept of defining the reference row/eye for
which the offset value must be zero.  Fangyi suggested we simplify the text and
remove the concept of a reference row for which the offset value must be zero.
He said we should just allow an offset to be provided for every row.  Walter had
no objections to this and removed the section describing the reference row and
the requirement that its offset value be zero.  Walter noted that there is still
a need to describe the nominal row/eye for statistical analysis.  Fangyi
suggested Walter move the definition of nominal (row n/2 if n is even, (n-1)/2
if n is odd) into the statistical analysis paragraph in the Other Notes:
section.  Walter agreed.

bit_time formal parameter of the AMI_Init function:
Fangyi said the change in the definition of bit_time should be rewritten to make
it clearer that it is the symbol time if Modulation is set to PAM4 or
Modulation_Levels is > 2.  Walter agreed.  Arpad then suggested that bit_time
should be renamed symbol_time, since symbol_time is an accurate name for every
value of n, but bit_time is only true for NRZ.  Fangyi and Radek agreed that we
should consider renaming this formal parameter of AMI_Init to symbol_time.
(Note: The IBIS 7.0 specification uses the heading "Arguments:" when it
 describes the formal parameters of the AMI functions)

Walter said he would sent out BIRD213.1 draft 6, and he noted that draft 5 had not
been sent out to ATM.

- Curtis: Motion to adjourn.
- Randy: Second.
- Arpad: Thank you all for joining.

AR: Fangyi send BIRD211.3 to ATM.
AR: Randy to upload BIRD211.3 to the Open Forum website and send an email
    announcement.
AR: Walter to send BIRD213.1 draft 6 to ATM.

-------------
Next meeting: 31 August 2021 12:00pm PT
-------------

IBIS Interconnect SPICE Wish List:

1) Simulator directives
